<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> <style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style> </head> <body dir="ltr"> <div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof ContentPasted0"> > On Wed, May 03, 2023 at 03:50:59PM -0700, fei.yang@intel.com wrote: <div class="ContentPasted0">>> From: Fei Yang <fei.yang@intel.com></div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> Currently the KMD is using enum i915_cache_level to set caching policy for</div> <div class="ContentPasted0">>> buffer objects. This is flaky because the PAT index which really controls</div> <div class="ContentPasted0">>> the caching behavior in PTE has far more levels than what's defined in the</div> <div class="ContentPasted0">>> enum. In addition, the PAT index is platform dependent, having to translate</div> <div class="ContentPasted0">>> between i915_cache_level and PAT index is not reliable, and makes the code</div> <div class="ContentPasted0">>> more complicated.</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> From UMD's perspective there is also a necessity to set caching policy for</div> <div class="ContentPasted0">>> performance fine tuning. It's much easier for the UMD to directly use PAT</div> <div class="ContentPasted0">>> index because the behavior of each PAT index is clearly defined in Bspec.</div> <div class="ContentPasted0">>> Having the abstracted i915_cache_level sitting in between would only cause</div> <div class="ContentPasted0">>> more ambiguity.</div> <div class="ContentPasted0">></div> <div class="ContentPasted0">> It may be worth mentioning here that PAT is expected to work much like</div> <div class="ContentPasted0">> MOCS already works today --- the contract on the exact platform-specific</div> <div class="ContentPasted0">> meaning of each index is documented in the hardware spec and userspace</div> <div class="ContentPasted0">> is expected to select the index that exactly matches the behavior it</div> <div class="ContentPasted0">> desires.</div> <div><br class="ContentPasted0"> </div> <div class="ContentPasted0">will update.</div> <div><br class="ContentPasted0"> </div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> For these reasons this patch replaces i915_cache_level with PAT index. Also</div> <div class="ContentPasted0">>> note, the cache_level is not completely removed yet, because the KMD still</div> <div class="ContentPasted0">>> has the need of creating buffer objects with simple cache settings such as</div> <div class="ContentPasted0">>> cached, uncached, or writethrough. For such simple cases, using cache_level</div> <div class="ContentPasted0">>> would help simplify the code.</div> <div class="ContentPasted0">></div> <div class="ContentPasted0">> See my comment farther down, but the implementation of</div> <div class="ContentPasted0">> i915_gem_object_has_cache_level() seems a bit confusing and you may want</div> <div class="ContentPasted0">> to elaborate on it here.</div> <div class="ContentPasted0">></div> <div class="ContentPasted0">> Also somewhat confusing from a high-level skim is if/how we're still</div> <div class="ContentPasted0">> using obj->cache_coherent once userspace has taken direct control of the</div> <div class="ContentPasted0">> cache behavior. Some PAT indices give coherency and others don't (and</div> <div class="ContentPasted0">> the combinations will likely get more complicated on future platforms).</div> <div class="ContentPasted0">> If obj->cache_coherent is still being considered even once PAT indices</div> <div class="ContentPasted0">> are being controlled by userspace, I think we need some explanation of</div> <div class="ContentPasted0">> how that works in the commit message (and likely in the kerneldoc for</div> <div class="ContentPasted0">> that field too).</div> <div><br class="ContentPasted0"> </div> <div class="ContentPasted0">For the objects with pat_index set by user space, coherency is managed by</div> <div class="ContentPasted0">user space. Things like obj->cache_coherent and obj->cache_dirty are still</div> <div class="ContentPasted0">there for objects created by kernel.</div> <div><br class="ContentPasted0"> </div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> Cc: Chris Wilson <chris.p.wilson@linux.intel.com></div> <div class="ContentPasted0">>> Cc: Matt Roper <matthew.d.roper@intel.com></div> <div class="ContentPasted0">>> Signed-off-by: Fei Yang <fei.yang@intel.com></div> <div class="ContentPasted0">>> Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com></div> <div class="ContentPasted0">>> ---</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/display/intel_dpt.c | 12 +--</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gem/i915_gem_domain.c | 45 ++++++----</div> <div class="ContentPasted0">>> .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 10 ++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gem/i915_gem_mman.c | 3 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gem/i915_gem_object.c | 51 +++++++++++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gem/i915_gem_object.h | 4 +</div> <div class="ContentPasted0">>> .../gpu/drm/i915/gem/i915_gem_object_types.h | 25 +++++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 4 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c | 16 ++--</div> <div class="ContentPasted0">>> .../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +-</div> <div class="ContentPasted0">>> .../drm/i915/gem/selftests/i915_gem_migrate.c | 2 +-</div> <div class="ContentPasted0">>> .../drm/i915/gem/selftests/i915_gem_mman.c | 2 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 10 ++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 71 ++++++++--------</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/gen8_ppgtt.h | 3 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/intel_ggtt.c | 82 +++++++++----------</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/intel_gtt.h | 20 ++---</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/intel_migrate.c | 47 ++++++-----</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/intel_migrate.h | 13 ++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/intel_ppgtt.c | 6 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/selftest_migrate.c | 47 ++++++-----</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/selftest_reset.c | 8 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/selftest_timeline.c | 2 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/selftest_tlb.c | 4 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 10 ++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/i915_debugfs.c | 52 +++++++++---</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/i915_gem.c | 16 +++-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/i915_gpu_error.c | 8 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/i915_vma.c | 16 ++--</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/i915_vma.h | 2 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/i915_vma_types.h | 2 -</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/selftests/i915_gem.c | 5 +-</div> <div class="ContentPasted0">>> .../gpu/drm/i915/selftests/i915_gem_evict.c | 4 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 15 ++--</div> <div class="ContentPasted0">>> .../drm/i915/selftests/intel_memory_region.c | 4 +-</div> <div class="ContentPasted0">>> drivers/gpu/drm/i915/selftests/mock_gtt.c | 8 +-</div> <div class="ContentPasted0">>> 36 files changed, 391 insertions(+), 240 deletions(-)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/display/intel_dpt.c b/drivers/gpu/drm/i915/display/intel_dpt.c</div> <div class="ContentPasted0">>> index c5eacfdba1a5..7c5fddb203ba 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/display/intel_dpt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/display/intel_dpt.c</div> <div class="ContentPasted0">>> @@ -43,24 +43,24 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)</div> <div class="ContentPasted0">>> static void dpt_insert_page(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_dpt *dpt = i915_vm_to_dpt(vm);</div> <div class="ContentPasted0">>> gen8_pte_t __iomem *base = dpt->iomem;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> gen8_set_pte(base + offset / I915_GTT_PAGE_SIZE,</div> <div class="ContentPasted0">>> - vm->pte_encode(addr, level, flags));</div> <div class="ContentPasted0">>> + vm->pte_encode(addr, pat_index, flags));</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void dpt_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_dpt *dpt = i915_vm_to_dpt(vm);</div> <div class="ContentPasted0">>> gen8_pte_t __iomem *base = dpt->iomem;</div> <div class="ContentPasted0">>> - const gen8_pte_t pte_encode = vm->pte_encode(0, level, flags);</div> <div class="ContentPasted0">>> + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);</div> <div class="ContentPasted0">>> struct sgt_iter sgt_iter;</div> <div class="ContentPasted0">>> dma_addr_t addr;</div> <div class="ContentPasted0">>> int i;</div> <div class="ContentPasted0">>> @@ -83,7 +83,7 @@ static void dpt_clear_range(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> static void dpt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> u32 pte_flags;</div> <div class="ContentPasted0">>> @@ -98,7 +98,7 @@ static void dpt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> if (vma_res->bi.lmem)</div> <div class="ContentPasted0">>> pte_flags |= PTE_LM;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - vm->insert_entries(vm, vma_res, cache_level, pte_flags);</div> <div class="ContentPasted0">>> + vm->insert_entries(vm, vma_res, pat_index, pte_flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_domain.c b/drivers/gpu/drm/i915/gem/i915_gem_domain.c</div> <div class="ContentPasted0">>> index d2d5a24301b2..ae99b4be5918 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_domain.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_domain.c</div> <div class="ContentPasted0">>> @@ -27,8 +27,8 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)</div> <div class="ContentPasted0">>> if (IS_DGFX(i915))</div> <div class="ContentPasted0">>> return false;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - return !(obj->cache_level == I915_CACHE_NONE ||</div> <div class="ContentPasted0">>> - obj->cache_level == I915_CACHE_WT);</div> <div class="ContentPasted0">>> + return !(i915_gem_object_has_cache_level(obj, I915_CACHE_NONE) ||</div> <div class="ContentPasted0">>> + i915_gem_object_has_cache_level(obj, I915_CACHE_WT));</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> bool i915_gem_cpu_write_needs_clflush(struct drm_i915_gem_object *obj)</div> <div class="ContentPasted0">>> @@ -267,7 +267,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> int ret;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - if (obj->cache_level == cache_level)</div> <div class="ContentPasted0">>> + if (i915_gem_object_has_cache_level(obj, cache_level))</div> <div class="ContentPasted0">>> return 0;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ret = i915_gem_object_wait(obj,</div> <div class="ContentPasted0">>> @@ -278,10 +278,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> return ret;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* Always invalidate stale cachelines */</div> <div class="ContentPasted0">>> - if (obj->cache_level != cache_level) {</div> <div class="ContentPasted0">>> - i915_gem_object_set_cache_coherency(obj, cache_level);</div> <div class="ContentPasted0">>> - obj->cache_dirty = true;</div> <div class="ContentPasted0">>> - }</div> <div class="ContentPasted0">>> + i915_gem_object_set_cache_coherency(obj, cache_level);</div> <div class="ContentPasted0">>> + obj->cache_dirty = true;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* The cache-level will be applied when each vma is rebound. */</div> <div class="ContentPasted0">>> return i915_gem_object_unbind(obj,</div> <div class="ContentPasted0">>> @@ -306,20 +304,22 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data,</div> <div class="ContentPasted0">>> goto out;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - switch (obj->cache_level) {</div> <div class="ContentPasted0">>> - case I915_CACHE_LLC:</div> <div class="ContentPasted0">>> - case I915_CACHE_L3_LLC:</div> <div class="ContentPasted0">>> - args->caching = I915_CACHING_CACHED;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> + /*</div> <div class="ContentPasted0">>> + * This ioctl should be disabled for the objects with pat_index</div> <div class="ContentPasted0">>> + * set by user space.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> + if (obj->pat_set_by_user) {</div> <div class="ContentPasted0">>> + err = -EOPNOTSUPP;</div> <div class="ContentPasted0">>> + goto out;</div> <div class="ContentPasted0">>> + }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - case I915_CACHE_WT:</div> <div class="ContentPasted0">>> + if (i915_gem_object_has_cache_level(obj, I915_CACHE_LLC) ||</div> <div class="ContentPasted0">>> + i915_gem_object_has_cache_level(obj, I915_CACHE_L3_LLC))</div> <div class="ContentPasted0">>> + args->caching = I915_CACHING_CACHED;</div> <div class="ContentPasted0">>> + else if (i915_gem_object_has_cache_level(obj, I915_CACHE_WT))</div> <div class="ContentPasted0">>> args->caching = I915_CACHING_DISPLAY;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> -</div> <div class="ContentPasted0">>> - default:</div> <div class="ContentPasted0">>> + else</div> <div class="ContentPasted0">>> args->caching = I915_CACHING_NONE;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - }</div> <div class="ContentPasted0">>> out:</div> <div class="ContentPasted0">>> rcu_read_unlock();</div> <div class="ContentPasted0">>> return err;</div> <div class="ContentPasted0">>> @@ -364,6 +364,15 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,</div> <div class="ContentPasted0">>> if (!obj)</div> <div class="ContentPasted0">>> return -ENOENT;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> + /*</div> <div class="ContentPasted0">>> + * This ioctl should be disabled for the objects with pat_index</div> <div class="ContentPasted0">>> + * set by user space.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> + if (obj->pat_set_by_user) {</div> <div class="ContentPasted0">>> + ret = -EOPNOTSUPP;</div> <div class="ContentPasted0">>> + goto out;</div> <div class="ContentPasted0">>> + }</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> /*</div> <div class="ContentPasted0">>> * The caching mode of proxy object is handled by its generator, and</div> <div class="ContentPasted0">>> * not allowed to be changed by userspace.</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c</div> <div class="ContentPasted0">>> index 3aeede6aee4d..d42915516636 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c</div> <div class="ContentPasted0">>> @@ -642,7 +642,7 @@ static inline int use_cpu_reloc(const struct reloc_cache *cache,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return (cache->has_llc ||</div> <div class="ContentPasted0">>> obj->cache_dirty ||</div> <div class="ContentPasted0">>> - obj->cache_level != I915_CACHE_NONE);</div> <div class="ContentPasted0">>> + !i915_gem_object_has_cache_level(obj, I915_CACHE_NONE));</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static int eb_reserve_vma(struct i915_execbuffer *eb,</div> <div class="ContentPasted0">>> @@ -1323,8 +1323,10 @@ static void *reloc_iomap(struct i915_vma *batch,</div> <div class="ContentPasted0">>> offset = cache->node.start;</div> <div class="ContentPasted0">>> if (drm_mm_node_allocated(&cache->node)) {</div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm,</div> <div class="ContentPasted0">>> - i915_gem_object_get_dma_address(obj, page),</div> <div class="ContentPasted0">>> - offset, I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_object_get_dma_address(obj, page),</div> <div class="ContentPasted0">>> + offset,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(ggtt->vm.i915, I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> } else {</div> <div class="ContentPasted0">>> offset += page << PAGE_SHIFT;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -1464,7 +1466,7 @@ eb_relocate_entry(struct i915_execbuffer *eb,</div> <div class="ContentPasted0">>> reloc_cache_unmap(&eb->reloc_cache);</div> <div class="ContentPasted0">>> mutex_lock(&vma->vm->mutex);</div> <div class="ContentPasted0">>> err = i915_vma_bind(target->vma,</div> <div class="ContentPasted0">>> - target->vma->obj->cache_level,</div> <div class="ContentPasted0">>> + target->vma->obj->pat_index,</div> <div class="ContentPasted0">>> PIN_GLOBAL, NULL, NULL);</div> <div class="ContentPasted0">>> mutex_unlock(&vma->vm->mutex);</div> <div class="ContentPasted0">>> reloc_cache_remap(&eb->reloc_cache, ev->vma->obj);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c</div> <div class="ContentPasted0">>> index 3dbacdf0911a..50c30efa08a3 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c</div> <div class="ContentPasted0">>> @@ -383,7 +383,8 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* Access to snoopable pages through the GTT is incoherent. */</div> <div class="ContentPasted0">>> - if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(i915)) {</div> <div class="ContentPasted0">>> + if (!(i915_gem_object_has_cache_level(obj, I915_CACHE_NONE) ||</div> <div class="ContentPasted0">>> + HAS_LLC(i915))) {</div> <div class="ContentPasted0">>> ret = -EFAULT;</div> <div class="ContentPasted0">>> goto err_unpin;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c</div> <div class="ContentPasted0">>> index 8c70a0ec7d2f..46a19b099ec8 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_object.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c</div> <div class="ContentPasted0">>> @@ -54,6 +54,24 @@ unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915,</div> <div class="ContentPasted0">>> return INTEL_INFO(i915)->cachelevel_to_pat[level];</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> +bool i915_gem_object_has_cache_level(const struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> + enum i915_cache_level lvl)</div> <div class="ContentPasted0">>> +{</div> <div class="ContentPasted0">>> + /*</div> <div class="ContentPasted0">>> + * In case the pat_index is set by user space, this kernel mode</div> <div class="ContentPasted0">>> + * driver should leave the coherency to be managed by user space,</div> <div class="ContentPasted0">>> + * simply return true here.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> + if (obj->pat_set_by_user)</div> <div class="ContentPasted0">>> + return true;</div> <div class="ContentPasted0">></div> <div class="ContentPasted0">> This function gets called in a few different places; if the question is</div> <div class="ContentPasted0">> "does the object have specific cache behavior XX" then universally</div> <div class="ContentPasted0">> returning "yes" if the user has taken direct control of PAT seems</div> <div class="ContentPasted0">> confusing, especially since we're not looking at what cache level was</div> <div class="ContentPasted0">> being queried nor what PAT index was selected by userspace. A return</div> <div class="ContentPasted0">> value of 'true' here could be correct in some situations but not others,</div> <div class="ContentPasted0">> depending on how the caller intends to use that information. I assume</div> <div class="ContentPasted0">> you've gone through all the places this function gets called and ensured</div> <div class="ContentPasted0">> that a return value of true results in the expected behavior; you may</div> <div class="ContentPasted0">> want to elaborate on that in the commit message.</div> <div><br class="ContentPasted0"> </div> <div class="ContentPasted0">There is a patch later in the series disabling {set|get}_caching ioctl for</div> <div class="ContentPasted0">objects with pat_index set by userspace. With that the logic here would</div> <div class="ContentPasted0">make sure kernel won't touch the cache settings for these objects. There</div> <div class="ContentPasted0">is one caller vm_fault_gtt() might be a bit sensitive though, CI doesn't</div> <div class="ContentPasted0">report any problem here, but if further changes in this part of the code</div> <div class="ContentPasted0">were made, the logic here might need to be refined.</div> <div><br class="ContentPasted0"> </div> <div class="ContentPasted0">> Matt</div> <div class="ContentPasted0">></div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + /*</div> <div class="ContentPasted0">>> + * Otherwise the pat_index should have been converted from cache_level</div> <div class="ContentPasted0">>> + * so that the following comparison is valid.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> + return obj->pat_index == i915_gem_get_pat_index(obj_to_i915(obj), lvl);</div> <div class="ContentPasted0">>> +}</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> struct drm_i915_gem_object *i915_gem_object_alloc(void)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct drm_i915_gem_object *obj;</div> <div class="ContentPasted0">>> @@ -133,7 +151,7 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct drm_i915_private *i915 = to_i915(obj->base.dev);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - obj->cache_level = cache_level;</div> <div class="ContentPasted0">>> + obj->pat_index = i915_gem_get_pat_index(i915, cache_level);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (cache_level != I915_CACHE_NONE)</div> <div class="ContentPasted0">>> obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |</div> <div class="ContentPasted0">>> @@ -148,6 +166,37 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> !IS_DGFX(i915);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> +/**</div> <div class="ContentPasted0">>> + * i915_gem_object_set_pat_index - set PAT index to be used in PTE encode</div> <div class="ContentPasted0">>> + * @obj: #drm_i915_gem_object</div> <div class="ContentPasted0">>> + * @pat_index: PAT index</div> <div class="ContentPasted0">>> + *</div> <div class="ContentPasted0">>> + * This is a clone of i915_gem_object_set_cache_coherency taking pat index</div> <div class="ContentPasted0">>> + * instead of cache_level as its second argument.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> +void i915_gem_object_set_pat_index(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> + unsigned int pat_index)</div> <div class="ContentPasted0">>> +{</div> <div class="ContentPasted0">>> + struct drm_i915_private *i915 = to_i915(obj->base.dev);</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (obj->pat_index == pat_index)</div> <div class="ContentPasted0">>> + return;</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + obj->pat_index = pat_index;</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (pat_index != i915_gem_get_pat_index(i915, I915_CACHE_NONE))</div> <div class="ContentPasted0">>> + obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |</div> <div class="ContentPasted0">>> + I915_BO_CACHE_COHERENT_FOR_WRITE);</div> <div class="ContentPasted0">>> + else if (HAS_LLC(i915))</div> <div class="ContentPasted0">>> + obj->cache_coherent = I915_BO_CACHE_COHERENT_FOR_READ;</div> <div class="ContentPasted0">>> + else</div> <div class="ContentPasted0">>> + obj->cache_coherent = 0;</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + obj->cache_dirty =</div> <div class="ContentPasted0">>> + !(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE) &&</div> <div class="ContentPasted0">>> + !IS_DGFX(i915);</div> <div class="ContentPasted0">>> +}</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct drm_i915_private *i915 = to_i915(obj->base.dev);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h</div> <div class="ContentPasted0">>> index 4c92e17b4337..6f00aab10015 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_object.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h</div> <div class="ContentPasted0">>> @@ -34,6 +34,8 @@ static inline bool i915_gem_object_size_2big(u64 size)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> unsigned int i915_gem_get_pat_index(struct drm_i915_private *i915,</div> <div class="ContentPasted0">>> enum i915_cache_level level);</div> <div class="ContentPasted0">>> +bool i915_gem_object_has_cache_level(const struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> + enum i915_cache_level lvl);</div> <div class="ContentPasted0">>> void i915_gem_init__objects(struct drm_i915_private *i915);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> void i915_objects_module_exit(void);</div> <div class="ContentPasted0">>> @@ -764,6 +766,8 @@ bool i915_gem_object_has_unknown_state(struct drm_i915_gem_object *obj);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> unsigned int cache_level);</div> <div class="ContentPasted0">>> +void i915_gem_object_set_pat_index(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> + unsigned int pat_index);</div> <div class="ContentPasted0">>> bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj);</div> <div class="ContentPasted0">>> void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj);</div> <div class="ContentPasted0">>> void i915_gem_object_flush_if_display_locked(struct drm_i915_gem_object *obj);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h</div> <div class="ContentPasted0">>> index 41b35abccf88..e19151f6e279 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h</div> <div class="ContentPasted0">>> @@ -358,12 +358,29 @@ struct drm_i915_gem_object {</div> <div class="ContentPasted0">>> #define I915_BO_FLAG_STRUCT_PAGE BIT(0) /* Object backed by struct pages */</div> <div class="ContentPasted0">>> #define I915_BO_FLAG_IOMEM BIT(1) /* Object backed by IO memory */</div> <div class="ContentPasted0">>> /**</div> <div class="ContentPasted0">>> - * @cache_level: The desired GTT caching level.</div> <div class="ContentPasted0">>> + * @pat_index: The desired PAT index.</div> <div class="ContentPasted0">>> + *</div> <div class="ContentPasted0">>> + * See hardware specification for valid PAT indices for each platform.</div> <div class="ContentPasted0">>> + * This field used to contain a value of enum i915_cache_level. It's</div> <div class="ContentPasted0">>> + * changed to an unsigned int because PAT indices are being used by</div> <div class="ContentPasted0">>> + * both UMD and KMD for caching policy control after GEN12.</div> <div class="ContentPasted0">>> + * For backward compatibility, this field will continue to contain</div> <div class="ContentPasted0">>> + * value of i915_cache_level for pre-GEN12 platforms so that the PTE</div> <div class="ContentPasted0">>> + * encode functions for these legacy platforms can stay the same.</div> <div class="ContentPasted0">>> + * In the meantime platform specific tables are created to translate</div> <div class="ContentPasted0">>> + * i915_cache_level into pat index, for more details check the macros</div> <div class="ContentPasted0">>> + * defined i915/i915_pci.c, e.g. PVC_CACHELEVEL.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> + unsigned int pat_index:6;</div> <div class="ContentPasted0">>> + /**</div> <div class="ContentPasted0">>> + * @pat_set_by_user: Indicate whether pat_index is set by user space</div> <div class="ContentPasted0">>> *</div> <div class="ContentPasted0">>> - * See enum i915_cache_level for possible values, along with what</div> <div class="ContentPasted0">>> - * each does.</div> <div class="ContentPasted0">>> + * This field is set to false by default, only set to true if the</div> <div class="ContentPasted0">>> + * pat_index is set by user space. By design, user space is capable of</div> <div class="ContentPasted0">>> + * managing caching behavior by setting pat_index, in which case this</div> <div class="ContentPasted0">>> + * kernel mode driver should never touch the pat_index.</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> - unsigned int cache_level:3;</div> <div class="ContentPasted0">>> + unsigned int pat_set_by_user;</div> <div class="ContentPasted0">>> /**</div> <div class="ContentPasted0">>> * @cache_coherent:</div> <div class="ContentPasted0">>> *</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c</div> <div class="ContentPasted0">>> index ee492d823f1b..3b094d36a0b0 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c</div> <div class="ContentPasted0">>> @@ -565,7 +565,9 @@ static void dbg_poison(struct i915_ggtt *ggtt,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm, addr,</div> <div class="ContentPasted0">>> ggtt->error_capture.start,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(ggtt->vm.i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> mb();</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> s = io_mapping_map_wc(&ggtt->iomap,</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c b/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c</div> <div class="ContentPasted0">>> index 69eb20ed4d47..e40761e13c2a 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm_move.c</div> <div class="ContentPasted0">>> @@ -214,7 +214,8 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> intel_engine_pm_get(to_gt(i915)->migrate.context->engine);</div> <div class="ContentPasted0">>> ret = intel_context_migrate_clear(to_gt(i915)->migrate.context, deps,</div> <div class="ContentPasted0">>> - dst_st->sgl, dst_level,</div> <div class="ContentPasted0">>> + dst_st->sgl,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915, dst_level),</div> <div class="ContentPasted0">>> i915_ttm_gtt_binds_lmem(dst_mem),</div> <div class="ContentPasted0">>> 0, &rq);</div> <div class="ContentPasted0">>> } else {</div> <div class="ContentPasted0">>> @@ -227,12 +228,13 @@ static struct dma_fence *i915_ttm_accel_move(struct ttm_buffer_object *bo,</div> <div class="ContentPasted0">>> src_level = i915_ttm_cache_level(i915, bo->resource, src_ttm);</div> <div class="ContentPasted0">>> intel_engine_pm_get(to_gt(i915)->migrate.context->engine);</div> <div class="ContentPasted0">>> ret = intel_context_migrate_copy(to_gt(i915)->migrate.context,</div> <div class="ContentPasted0">>> - deps, src_rsgt->table.sgl,</div> <div class="ContentPasted0">>> - src_level,</div> <div class="ContentPasted0">>> - i915_ttm_gtt_binds_lmem(bo->resource),</div> <div class="ContentPasted0">>> - dst_st->sgl, dst_level,</div> <div class="ContentPasted0">>> - i915_ttm_gtt_binds_lmem(dst_mem),</div> <div class="ContentPasted0">>> - &rq);</div> <div class="ContentPasted0">>> + deps, src_rsgt->table.sgl,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915, src_level),</div> <div class="ContentPasted0">>> + i915_ttm_gtt_binds_lmem(bo->resource),</div> <div class="ContentPasted0">>> + dst_st->sgl,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915, dst_level),</div> <div class="ContentPasted0">>> + i915_ttm_gtt_binds_lmem(dst_mem),</div> <div class="ContentPasted0">>> + &rq);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> i915_refct_sgt_put(src_rsgt);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c</div> <div class="ContentPasted0">>> index cb5863f37f9d..763df437c20c 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c</div> <div class="ContentPasted0">>> @@ -354,7 +354,7 @@ fake_huge_pages_object(struct drm_i915_private *i915, u64 size, bool single)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> obj->write_domain = I915_GEM_DOMAIN_CPU;</div> <div class="ContentPasted0">>> obj->read_domains = I915_GEM_DOMAIN_CPU;</div> <div class="ContentPasted0">>> - obj->cache_level = I915_CACHE_NONE;</div> <div class="ContentPasted0">>> + obj->pat_index = i915_gem_get_pat_index(i915, I915_CACHE_NONE);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return obj;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c</div> <div class="ContentPasted0">>> index fe6c37fd7859..a93a90b15907 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_migrate.c</div> <div class="ContentPasted0">>> @@ -219,7 +219,7 @@ static int __igt_lmem_pages_migrate(struct intel_gt *gt,</div> <div class="ContentPasted0">>> continue;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = intel_migrate_clear(&gt->migrate, &ww, deps,</div> <div class="ContentPasted0">>> - obj->mm.pages->sgl, obj->cache_level,</div> <div class="ContentPasted0">>> + obj->mm.pages->sgl, obj->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(obj),</div> <div class="ContentPasted0">>> 0xdeadbeaf, &rq);</div> <div class="ContentPasted0">>> if (rq) {</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c</div> <div class="ContentPasted0">>> index 56279908ed30..a93d8f9f8bc1 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c</div> <div class="ContentPasted0">>> @@ -1222,7 +1222,7 @@ static int __igt_mmap_migrate(struct intel_memory_region **placements,</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = intel_context_migrate_clear(to_gt(i915)->migrate.context, NULL,</div> <div class="ContentPasted0">>> - obj->mm.pages->sgl, obj->cache_level,</div> <div class="ContentPasted0">>> + obj->mm.pages->sgl, obj->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(obj),</div> <div class="ContentPasted0">>> expand32(POISON_INUSE), &rq);</div> <div class="ContentPasted0">>> i915_gem_object_unpin_pages(obj);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c</div> <div class="ContentPasted0">>> index 5aaacc53fa4c..c2bdc133c89a 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c</div> <div class="ContentPasted0">>> @@ -109,7 +109,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);</div> <div class="ContentPasted0">>> @@ -117,7 +117,7 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> unsigned int first_entry = vma_res->start / I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> unsigned int act_pt = first_entry / GEN6_PTES;</div> <div class="ContentPasted0">>> unsigned int act_pte = first_entry % GEN6_PTES;</div> <div class="ContentPasted0">>> - const u32 pte_encode = vm->pte_encode(0, cache_level, flags);</div> <div class="ContentPasted0">>> + const u32 pte_encode = vm->pte_encode(0, pat_index, flags);</div> <div class="ContentPasted0">>> struct sgt_dma iter = sgt_dma(vma_res);</div> <div class="ContentPasted0">>> gen6_pte_t *vaddr;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -227,7 +227,9 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vm->scratch[0]->encode =</div> <div class="ContentPasted0">>> vm->pte_encode(px_dma(vm->scratch[0]),</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, PTE_READ_ONLY);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + PTE_READ_ONLY);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vm->scratch[1] = vm->alloc_pt_dma(vm, I915_GTT_PAGE_SIZE_4K);</div> <div class="ContentPasted0">>> if (IS_ERR(vm->scratch[1])) {</div> <div class="ContentPasted0">>> @@ -278,7 +280,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)</div> <div class="ContentPasted0">>> static void pd_vma_bind(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 unused)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c</div> <div class="ContentPasted0">>> index bb6998d67133..f2334a713c4e 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c</div> <div class="ContentPasted0">>> @@ -56,7 +56,7 @@ static u64 gen8_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static u64 mtl_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> gen8_pte_t pte = addr | GEN8_PAGE_PRESENT | GEN8_PAGE_RW;</div> <div class="ContentPasted0">>> @@ -67,24 +67,17 @@ static u64 mtl_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> if (flags & PTE_LM)</div> <div class="ContentPasted0">>> pte |= GEN12_PPGTT_PTE_LM;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - switch (level) {</div> <div class="ContentPasted0">>> - case I915_CACHE_NONE:</div> <div class="ContentPasted0">>> - pte |= GEN12_PPGTT_PTE_PAT1;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - case I915_CACHE_LLC:</div> <div class="ContentPasted0">>> - case I915_CACHE_L3_LLC:</div> <div class="ContentPasted0">>> - pte |= GEN12_PPGTT_PTE_PAT0 | GEN12_PPGTT_PTE_PAT1;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - case I915_CACHE_WT:</div> <div class="ContentPasted0">>> + if (pat_index & BIT(0))</div> <div class="ContentPasted0">>> pte |= GEN12_PPGTT_PTE_PAT0;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - default:</div> <div class="ContentPasted0">>> - /* This should never happen. Added to deal with the compile</div> <div class="ContentPasted0">>> - * error due to the addition of I915_MAX_CACHE_LEVEL. Will</div> <div class="ContentPasted0">>> - * be removed by the pat_index patch.</div> <div class="ContentPasted0">>> - */</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - }</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (pat_index & BIT(1))</div> <div class="ContentPasted0">>> + pte |= GEN12_PPGTT_PTE_PAT1;</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (pat_index & BIT(2))</div> <div class="ContentPasted0">>> + pte |= GEN12_PPGTT_PTE_PAT2;</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (pat_index & BIT(3))</div> <div class="ContentPasted0">>> + pte |= MTL_PPGTT_PTE_PAT3;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return pte;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -457,11 +450,11 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,</div> <div class="ContentPasted0">>> struct i915_page_directory *pdp,</div> <div class="ContentPasted0">>> struct sgt_dma *iter,</div> <div class="ContentPasted0">>> u64 idx,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_page_directory *pd;</div> <div class="ContentPasted0">>> - const gen8_pte_t pte_encode = ppgtt->vm.pte_encode(0, cache_level, flags);</div> <div class="ContentPasted0">>> + const gen8_pte_t pte_encode = ppgtt->vm.pte_encode(0, pat_index, flags);</div> <div class="ContentPasted0">>> gen8_pte_t *vaddr;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> pd = i915_pd_entry(pdp, gen8_pd_index(idx, 2));</div> <div class="ContentPasted0">>> @@ -504,10 +497,10 @@ static void</div> <div class="ContentPasted0">>> xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> struct sgt_dma *iter,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags);</div> <div class="ContentPasted0">>> + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);</div> <div class="ContentPasted0">>> unsigned int rem = sg_dma_len(iter->sg);</div> <div class="ContentPasted0">>> u64 start = vma_res->start;</div> <div class="ContentPasted0">>> u64 end = start + vma_res->vma_size;</div> <div class="ContentPasted0">>> @@ -612,10 +605,10 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> struct sgt_dma *iter,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> - const gen8_pte_t pte_encode = vm->pte_encode(0, cache_level, flags);</div> <div class="ContentPasted0">>> + const gen8_pte_t pte_encode = vm->pte_encode(0, pat_index, flags);</div> <div class="ContentPasted0">>> unsigned int rem = sg_dma_len(iter->sg);</div> <div class="ContentPasted0">>> u64 start = vma_res->start;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -735,7 +728,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void gen8_ppgtt_insert(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ppgtt * const ppgtt = i915_vm_to_ppgtt(vm);</div> <div class="ContentPasted0">>> @@ -743,9 +736,9 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (vma_res->bi.page_sizes.sg > I915_GTT_PAGE_SIZE) {</div> <div class="ContentPasted0">>> if (GRAPHICS_VER_FULL(vm->i915) >= IP_VER(12, 50))</div> <div class="ContentPasted0">>> - xehpsdv_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags);</div> <div class="ContentPasted0">>> + xehpsdv_ppgtt_insert_huge(vm, vma_res, &iter, pat_index, flags);</div> <div class="ContentPasted0">>> else</div> <div class="ContentPasted0">>> - gen8_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags);</div> <div class="ContentPasted0">>> + gen8_ppgtt_insert_huge(vm, vma_res, &iter, pat_index, flags);</div> <div class="ContentPasted0">>> } else {</div> <div class="ContentPasted0">>> u64 idx = vma_res->start >> GEN8_PTE_SHIFT;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -754,7 +747,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> gen8_pdp_for_page_index(vm, idx);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> idx = gen8_ppgtt_insert_pte(ppgtt, pdp, &iter, idx,</div> <div class="ContentPasted0">>> - cache_level, flags);</div> <div class="ContentPasted0">>> + pat_index, flags);</div> <div class="ContentPasted0">>> } while (idx);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> @@ -764,7 +757,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> u64 idx = offset >> GEN8_PTE_SHIFT;</div> <div class="ContentPasted0">>> @@ -778,14 +771,14 @@ static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> GEM_BUG_ON(pt->is_compact);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vaddr = px_vaddr(pt);</div> <div class="ContentPasted0">>> - vaddr[gen8_pd_index(idx, 0)] = vm->pte_encode(addr, level, flags);</div> <div class="ContentPasted0">>> + vaddr[gen8_pd_index(idx, 0)] = vm->pte_encode(addr, pat_index, flags);</div> <div class="ContentPasted0">>> drm_clflush_virt_range(&vaddr[gen8_pd_index(idx, 0)], sizeof(*vaddr));</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> u64 idx = offset >> GEN8_PTE_SHIFT;</div> <div class="ContentPasted0">>> @@ -808,20 +801,20 @@ static void __xehpsdv_ppgtt_insert_entry_lm(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vaddr = px_vaddr(pt);</div> <div class="ContentPasted0">>> - vaddr[gen8_pd_index(idx, 0) / 16] = vm->pte_encode(addr, level, flags);</div> <div class="ContentPasted0">>> + vaddr[gen8_pd_index(idx, 0) / 16] = vm->pte_encode(addr, pat_index, flags);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void xehpsdv_ppgtt_insert_entry(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> if (flags & PTE_LM)</div> <div class="ContentPasted0">>> return __xehpsdv_ppgtt_insert_entry_lm(vm, addr, offset,</div> <div class="ContentPasted0">>> - level, flags);</div> <div class="ContentPasted0">>> + pat_index, flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - return gen8_ppgtt_insert_entry(vm, addr, offset, level, flags);</div> <div class="ContentPasted0">>> + return gen8_ppgtt_insert_entry(vm, addr, offset, pat_index, flags);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static int gen8_init_scratch(struct i915_address_space *vm)</div> <div class="ContentPasted0">>> @@ -856,7 +849,9 @@ static int gen8_init_scratch(struct i915_address_space *vm)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vm->scratch[0]->encode =</div> <div class="ContentPasted0">>> vm->pte_encode(px_dma(vm->scratch[0]),</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, pte_flags);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + pte_flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> for (i = 1; i <= vm->top; i++) {</div> <div class="ContentPasted0">>> struct drm_i915_gem_object *obj;</div> <div class="ContentPasted0">>> @@ -874,7 +869,9 @@ static int gen8_init_scratch(struct i915_address_space *vm)</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> fill_px(obj, vm->scratch[i - 1]->encode);</div> <div class="ContentPasted0">>> - obj->encode = gen8_pde_encode(px_dma(obj), I915_CACHE_NONE);</div> <div class="ContentPasted0">>> + obj->encode = gen8_pde_encode(px_dma(obj),</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE));</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vm->scratch[i] = obj;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/gen8_ppgtt.h b/drivers/gpu/drm/i915/gt/gen8_ppgtt.h</div> <div class="ContentPasted0">>> index f541d19264b4..19c635441642 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/gen8_ppgtt.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/gen8_ppgtt.h</div> <div class="ContentPasted0">>> @@ -10,13 +10,12 @@</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> struct i915_address_space;</div> <div class="ContentPasted0">>> struct intel_gt;</div> <div class="ContentPasted0">>> -enum i915_cache_level;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> struct i915_ppgtt *gen8_ppgtt_create(struct intel_gt *gt,</div> <div class="ContentPasted0">>> unsigned long lmem_pt_obj_flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> u64 gen8_ggtt_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> #endif</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c</div> <div class="ContentPasted0">>> index c8390d03fce2..2a7942fac798 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/intel_ggtt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c</div> <div class="ContentPasted0">>> @@ -221,7 +221,7 @@ static void guc_ggtt_invalidate(struct i915_ggtt *ggtt)</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static u64 mtl_ggtt_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> gen8_pte_t pte = addr | GEN8_PAGE_PRESENT;</div> <div class="ContentPasted0">>> @@ -231,30 +231,17 @@ static u64 mtl_ggtt_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> if (flags & PTE_LM)</div> <div class="ContentPasted0">>> pte |= GEN12_GGTT_PTE_LM;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - switch (level) {</div> <div class="ContentPasted0">>> - case I915_CACHE_NONE:</div> <div class="ContentPasted0">>> - pte |= MTL_GGTT_PTE_PAT1;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - case I915_CACHE_LLC:</div> <div class="ContentPasted0">>> - case I915_CACHE_L3_LLC:</div> <div class="ContentPasted0">>> - pte |= MTL_GGTT_PTE_PAT0 | MTL_GGTT_PTE_PAT1;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - case I915_CACHE_WT:</div> <div class="ContentPasted0">>> + if (pat_index & BIT(0))</div> <div class="ContentPasted0">>> pte |= MTL_GGTT_PTE_PAT0;</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - default:</div> <div class="ContentPasted0">>> - /* This should never happen. Added to deal with the compile</div> <div class="ContentPasted0">>> - * error due to the addition of I915_MAX_CACHE_LEVEL. Will</div> <div class="ContentPasted0">>> - * be removed by the pat_index patch.</div> <div class="ContentPasted0">>> - */</div> <div class="ContentPasted0">>> - break;</div> <div class="ContentPasted0">>> - }</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (pat_index & BIT(1))</div> <div class="ContentPasted0">>> + pte |= MTL_GGTT_PTE_PAT1;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return pte;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> u64 gen8_ggtt_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> gen8_pte_t pte = addr | GEN8_PAGE_PRESENT;</div> <div class="ContentPasted0">>> @@ -273,25 +260,25 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)</div> <div class="ContentPasted0">>> static void gen8_ggtt_insert_page(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);</div> <div class="ContentPasted0">>> gen8_pte_t __iomem *pte =</div> <div class="ContentPasted0">>> (gen8_pte_t __iomem *)ggtt->gsm + offset / I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - gen8_set_pte(pte, ggtt->vm.pte_encode(addr, level, flags));</div> <div class="ContentPasted0">>> + gen8_set_pte(pte, ggtt->vm.pte_encode(addr, pat_index, flags));</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->invalidate(ggtt);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void gen8_ggtt_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);</div> <div class="ContentPasted0">>> - const gen8_pte_t pte_encode = ggtt->vm.pte_encode(0, level, flags);</div> <div class="ContentPasted0">>> + const gen8_pte_t pte_encode = ggtt->vm.pte_encode(0, pat_index, flags);</div> <div class="ContentPasted0">>> gen8_pte_t __iomem *gte;</div> <div class="ContentPasted0">>> gen8_pte_t __iomem *end;</div> <div class="ContentPasted0">>> struct sgt_iter iter;</div> <div class="ContentPasted0">>> @@ -348,14 +335,14 @@ static void gen8_ggtt_clear_range(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> static void gen6_ggtt_insert_page(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);</div> <div class="ContentPasted0">>> gen6_pte_t __iomem *pte =</div> <div class="ContentPasted0">>> (gen6_pte_t __iomem *)ggtt->gsm + offset / I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - iowrite32(vm->pte_encode(addr, level, flags), pte);</div> <div class="ContentPasted0">>> + iowrite32(vm->pte_encode(addr, pat_index, flags), pte);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->invalidate(ggtt);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -368,7 +355,7 @@ static void gen6_ggtt_insert_page(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> static void gen6_ggtt_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);</div> <div class="ContentPasted0">>> @@ -385,7 +372,7 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> iowrite32(vm->scratch[0]->encode, gte++);</div> <div class="ContentPasted0">>> end += (vma_res->node_size + vma_res->guard) / I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> for_each_sgt_daddr(addr, iter, vma_res->bi.pages)</div> <div class="ContentPasted0">>> - iowrite32(vm->pte_encode(addr, level, flags), gte++);</div> <div class="ContentPasted0">>> + iowrite32(vm->pte_encode(addr, pat_index, flags), gte++);</div> <div class="ContentPasted0">>> GEM_BUG_ON(gte > end);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* Fill the allocated but "unused" space beyond the end of the buffer */</div> <div class="ContentPasted0">>> @@ -420,14 +407,15 @@ struct insert_page {</div> <div class="ContentPasted0">>> struct i915_address_space *vm;</div> <div class="ContentPasted0">>> dma_addr_t addr;</div> <div class="ContentPasted0">>> u64 offset;</div> <div class="ContentPasted0">>> - enum i915_cache_level level;</div> <div class="ContentPasted0">>> + unsigned int pat_index;</div> <div class="ContentPasted0">>> };</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static int bxt_vtd_ggtt_insert_page__cb(void *_arg)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct insert_page *arg = _arg;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - gen8_ggtt_insert_page(arg->vm, arg->addr, arg->offset, arg->level, 0);</div> <div class="ContentPasted0">>> + gen8_ggtt_insert_page(arg->vm, arg->addr, arg->offset,</div> <div class="ContentPasted0">>> + arg->pat_index, 0);</div> <div class="ContentPasted0">>> bxt_vtd_ggtt_wa(arg->vm);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return 0;</div> <div class="ContentPasted0">>> @@ -436,10 +424,10 @@ static int bxt_vtd_ggtt_insert_page__cb(void *_arg)</div> <div class="ContentPasted0">>> static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 unused)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> - struct insert_page arg = { vm, addr, offset, level };</div> <div class="ContentPasted0">>> + struct insert_page arg = { vm, addr, offset, pat_index };</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> stop_machine(bxt_vtd_ggtt_insert_page__cb, &arg, NULL);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -447,7 +435,7 @@ static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct insert_entries {</div> <div class="ContentPasted0">>> struct i915_address_space *vm;</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res;</div> <div class="ContentPasted0">>> - enum i915_cache_level level;</div> <div class="ContentPasted0">>> + unsigned int pat_index;</div> <div class="ContentPasted0">>> u32 flags;</div> <div class="ContentPasted0">>> };</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -455,7 +443,8 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct insert_entries *arg = _arg;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - gen8_ggtt_insert_entries(arg->vm, arg->vma_res, arg->level, arg->flags);</div> <div class="ContentPasted0">>> + gen8_ggtt_insert_entries(arg->vm, arg->vma_res,</div> <div class="ContentPasted0">>> + arg->pat_index, arg->flags);</div> <div class="ContentPasted0">>> bxt_vtd_ggtt_wa(arg->vm);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return 0;</div> <div class="ContentPasted0">>> @@ -463,10 +452,10 @@ static int bxt_vtd_ggtt_insert_entries__cb(void *_arg)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void bxt_vtd_ggtt_insert_entries__BKL(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> - struct insert_entries arg = { vm, vma_res, level, flags };</div> <div class="ContentPasted0">>> + struct insert_entries arg = { vm, vma_res, pat_index, flags };</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> stop_machine(bxt_vtd_ggtt_insert_entries__cb, &arg, NULL);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -495,7 +484,7 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> void intel_ggtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> u32 pte_flags;</div> <div class="ContentPasted0">>> @@ -512,7 +501,7 @@ void intel_ggtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> if (vma_res->bi.lmem)</div> <div class="ContentPasted0">>> pte_flags |= PTE_LM;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - vm->insert_entries(vm, vma_res, cache_level, pte_flags);</div> <div class="ContentPasted0">>> + vm->insert_entries(vm, vma_res, pat_index, pte_flags);</div> <div class="ContentPasted0">>> vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -661,7 +650,7 @@ static int init_ggtt(struct i915_ggtt *ggtt)</div> <div class="ContentPasted0">>> static void aliasing_gtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> u32 pte_flags;</div> <div class="ContentPasted0">>> @@ -673,10 +662,10 @@ static void aliasing_gtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (flags & I915_VMA_LOCAL_BIND)</div> <div class="ContentPasted0">>> ppgtt_bind_vma(&i915_vm_to_ggtt(vm)->alias->vm,</div> <div class="ContentPasted0">>> - stash, vma_res, cache_level, flags);</div> <div class="ContentPasted0">>> + stash, vma_res, pat_index, flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (flags & I915_VMA_GLOBAL_BIND)</div> <div class="ContentPasted0">>> - vm->insert_entries(vm, vma_res, cache_level, pte_flags);</div> <div class="ContentPasted0">>> + vm->insert_entries(vm, vma_res, pat_index, pte_flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vma_res->bound_flags |= flags;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -933,7 +922,9 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->vm.scratch[0]->encode =</div> <div class="ContentPasted0">>> ggtt->vm.pte_encode(px_dma(ggtt->vm.scratch[0]),</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, pte_flags);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + pte_flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return 0;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -1022,6 +1013,11 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)</div> <div class="ContentPasted0">>> return ggtt_probe_common(ggtt, size);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> +/*</div> <div class="ContentPasted0">>> + * For pre-gen8 platforms pat_index is the same as enum i915_cache_level,</div> <div class="ContentPasted0">>> + * so these PTE encode functions are left with using cache_level.</div> <div class="ContentPasted0">>> + * See translation table LEGACY_CACHELEVEL.</div> <div class="ContentPasted0">>> + */</div> <div class="ContentPasted0">>> static u64 snb_pte_encode(dma_addr_t addr,</div> <div class="ContentPasted0">>> enum i915_cache_level level,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> @@ -1302,7 +1298,9 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> vma->resource->bound_flags = 0;</div> <div class="ContentPasted0">>> vma->ops->bind_vma(vm, NULL, vma->resource,</div> <div class="ContentPasted0">>> - obj ? obj->cache_level : 0,</div> <div class="ContentPasted0">>> + obj ? obj->pat_index :</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> was_bound);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (obj) { /* only used during resume => exclusive access */</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h</div> <div class="ContentPasted0">>> index 1910683f03b4..3b86f66f64d3 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/intel_gtt.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/intel_gtt.h</div> <div class="ContentPasted0">>> @@ -163,8 +163,6 @@ typedef u64 gen8_pte_t;</div> <div class="ContentPasted0">>> #define MTL_3_COH_2W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 3)</div> <div class="ContentPasted0">>> #define MTL_2_COH_1W REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> -enum i915_cache_level;</div> <div class="ContentPasted0">>> -</div> <div class="ContentPasted0">>> struct drm_i915_gem_object;</div> <div class="ContentPasted0">>> struct i915_fence_reg;</div> <div class="ContentPasted0">>> struct i915_vma;</div> <div class="ContentPasted0">>> @@ -232,7 +230,7 @@ struct i915_vma_ops {</div> <div class="ContentPasted0">>> void (*bind_vma)(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> /*</div> <div class="ContentPasted0">>> * Unmap an object from an address space. This usually consists of</div> <div class="ContentPasted0">>> @@ -304,7 +302,7 @@ struct i915_address_space {</div> <div class="ContentPasted0">>> (*alloc_scratch_dma)(struct i915_address_space *vm, int sz);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> u64 (*pte_encode)(dma_addr_t addr,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags); /* Create a valid PTE */</div> <div class="ContentPasted0">>> #define PTE_READ_ONLY BIT(0)</div> <div class="ContentPasted0">>> #define PTE_LM BIT(1)</div> <div class="ContentPasted0">>> @@ -319,20 +317,20 @@ struct i915_address_space {</div> <div class="ContentPasted0">>> void (*insert_page)(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> void (*insert_entries)(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> void (*raw_insert_page)(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> void (*raw_insert_entries)(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> void (*cleanup)(struct i915_address_space *vm);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -579,7 +577,7 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,</div> <div class="ContentPasted0">>> void intel_ggtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> void intel_ggtt_unbind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res);</div> <div class="ContentPasted0">>> @@ -637,7 +635,7 @@ void</div> <div class="ContentPasted0">>> __set_pd_entry(struct i915_page_directory * const pd,</div> <div class="ContentPasted0">>> const unsigned short idx,</div> <div class="ContentPasted0">>> struct i915_page_table *pt,</div> <div class="ContentPasted0">>> - u64 (*encode)(const dma_addr_t, const enum i915_cache_level));</div> <div class="ContentPasted0">>> + u64 (*encode)(const dma_addr_t, const unsigned int pat_index));</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> #define set_pd_entry(pd, idx, to) \</div> <div class="ContentPasted0">>> __set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)</div> <div class="ContentPasted0">>> @@ -657,7 +655,7 @@ void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);</div> <div class="ContentPasted0">>> void ppgtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags);</div> <div class="ContentPasted0">>> void ppgtt_unbind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c</div> <div class="ContentPasted0">>> index 3f638f198796..117c3d05af3e 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/intel_migrate.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c</div> <div class="ContentPasted0">>> @@ -45,7 +45,9 @@ static void xehpsdv_toggle_pdes(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> * Insert a dummy PTE into every PT that will map to LMEM to ensure</div> <div class="ContentPasted0">>> * we have a correctly setup PDE structure for later use.</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> - vm->insert_page(vm, 0, d->offset, I915_CACHE_NONE, PTE_LM);</div> <div class="ContentPasted0">>> + vm->insert_page(vm, 0, d->offset,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915, I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + PTE_LM);</div> <div class="ContentPasted0">>> GEM_BUG_ON(!pt->is_compact);</div> <div class="ContentPasted0">>> d->offset += SZ_2M;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -63,7 +65,9 @@ static void xehpsdv_insert_pte(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> * alignment is 64K underneath for the pt, and we are careful</div> <div class="ContentPasted0">>> * not to access the space in the void.</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> - vm->insert_page(vm, px_dma(pt), d->offset, I915_CACHE_NONE, PTE_LM);</div> <div class="ContentPasted0">>> + vm->insert_page(vm, px_dma(pt), d->offset,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915, I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + PTE_LM);</div> <div class="ContentPasted0">>> d->offset += SZ_64K;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -73,7 +77,8 @@ static void insert_pte(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct insert_pte_data *d = data;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - vm->insert_page(vm, px_dma(pt), d->offset, I915_CACHE_NONE,</div> <div class="ContentPasted0">>> + vm->insert_page(vm, px_dma(pt), d->offset,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915, I915_CACHE_NONE),</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(pt->base) ? PTE_LM : 0);</div> <div class="ContentPasted0">>> d->offset += PAGE_SIZE;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -356,13 +361,13 @@ static int max_pte_pkt_size(struct i915_request *rq, int pkt)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static int emit_pte(struct i915_request *rq,</div> <div class="ContentPasted0">>> struct sgt_dma *it,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool is_lmem,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> int length)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> bool has_64K_pages = HAS_64K_PAGES(rq->engine->i915);</div> <div class="ContentPasted0">>> - const u64 encode = rq->context->vm->pte_encode(0, cache_level,</div> <div class="ContentPasted0">>> + const u64 encode = rq->context->vm->pte_encode(0, pat_index,</div> <div class="ContentPasted0">>> is_lmem ? PTE_LM : 0);</div> <div class="ContentPasted0">>> struct intel_ring *ring = rq->ring;</div> <div class="ContentPasted0">>> int pkt, dword_length;</div> <div class="ContentPasted0">>> @@ -673,17 +678,17 @@ int</div> <div class="ContentPasted0">>> intel_context_migrate_copy(struct intel_context *ce,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *src,</div> <div class="ContentPasted0">>> - enum i915_cache_level src_cache_level,</div> <div class="ContentPasted0">>> + unsigned int src_pat_index,</div> <div class="ContentPasted0">>> bool src_is_lmem,</div> <div class="ContentPasted0">>> struct scatterlist *dst,</div> <div class="ContentPasted0">>> - enum i915_cache_level dst_cache_level,</div> <div class="ContentPasted0">>> + unsigned int dst_pat_index,</div> <div class="ContentPasted0">>> bool dst_is_lmem,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> struct sgt_dma it_src = sg_sgt(src), it_dst = sg_sgt(dst), it_ccs;</div> <div class="ContentPasted0">>> struct drm_i915_private *i915 = ce->engine->i915;</div> <div class="ContentPasted0">>> u64 ccs_bytes_to_cpy = 0, bytes_to_cpy;</div> <div class="ContentPasted0">>> - enum i915_cache_level ccs_cache_level;</div> <div class="ContentPasted0">>> + unsigned int ccs_pat_index;</div> <div class="ContentPasted0">>> u32 src_offset, dst_offset;</div> <div class="ContentPasted0">>> u8 src_access, dst_access;</div> <div class="ContentPasted0">>> struct i915_request *rq;</div> <div class="ContentPasted0">>> @@ -707,12 +712,12 @@ intel_context_migrate_copy(struct intel_context *ce,</div> <div class="ContentPasted0">>> dst_sz = scatter_list_length(dst);</div> <div class="ContentPasted0">>> if (src_is_lmem) {</div> <div class="ContentPasted0">>> it_ccs = it_dst;</div> <div class="ContentPasted0">>> - ccs_cache_level = dst_cache_level;</div> <div class="ContentPasted0">>> + ccs_pat_index = dst_pat_index;</div> <div class="ContentPasted0">>> ccs_is_src = false;</div> <div class="ContentPasted0">>> } else if (dst_is_lmem) {</div> <div class="ContentPasted0">>> bytes_to_cpy = dst_sz;</div> <div class="ContentPasted0">>> it_ccs = it_src;</div> <div class="ContentPasted0">>> - ccs_cache_level = src_cache_level;</div> <div class="ContentPasted0">>> + ccs_pat_index = src_pat_index;</div> <div class="ContentPasted0">>> ccs_is_src = true;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -773,7 +778,7 @@ intel_context_migrate_copy(struct intel_context *ce,</div> <div class="ContentPasted0">>> src_sz = calculate_chunk_sz(i915, src_is_lmem,</div> <div class="ContentPasted0">>> bytes_to_cpy, ccs_bytes_to_cpy);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - len = emit_pte(rq, &it_src, src_cache_level, src_is_lmem,</div> <div class="ContentPasted0">>> + len = emit_pte(rq, &it_src, src_pat_index, src_is_lmem,</div> <div class="ContentPasted0">>> src_offset, src_sz);</div> <div class="ContentPasted0">>> if (!len) {</div> <div class="ContentPasted0">>> err = -EINVAL;</div> <div class="ContentPasted0">>> @@ -784,7 +789,7 @@ intel_context_migrate_copy(struct intel_context *ce,</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - err = emit_pte(rq, &it_dst, dst_cache_level, dst_is_lmem,</div> <div class="ContentPasted0">>> + err = emit_pte(rq, &it_dst, dst_pat_index, dst_is_lmem,</div> <div class="ContentPasted0">>> dst_offset, len);</div> <div class="ContentPasted0">>> if (err < 0)</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> @@ -811,7 +816,7 @@ intel_context_migrate_copy(struct intel_context *ce,</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ccs_sz = GET_CCS_BYTES(i915, len);</div> <div class="ContentPasted0">>> - err = emit_pte(rq, &it_ccs, ccs_cache_level, false,</div> <div class="ContentPasted0">>> + err = emit_pte(rq, &it_ccs, ccs_pat_index, false,</div> <div class="ContentPasted0">>> ccs_is_src ? src_offset : dst_offset,</div> <div class="ContentPasted0">>> ccs_sz);</div> <div class="ContentPasted0">>> if (err < 0)</div> <div class="ContentPasted0">>> @@ -979,7 +984,7 @@ int</div> <div class="ContentPasted0">>> intel_context_migrate_clear(struct intel_context *ce,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool is_lmem,</div> <div class="ContentPasted0">>> u32 value,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> @@ -1027,7 +1032,7 @@ intel_context_migrate_clear(struct intel_context *ce,</div> <div class="ContentPasted0">>> if (err)</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - len = emit_pte(rq, &it, cache_level, is_lmem, offset, CHUNK_SZ);</div> <div class="ContentPasted0">>> + len = emit_pte(rq, &it, pat_index, is_lmem, offset, CHUNK_SZ);</div> <div class="ContentPasted0">>> if (len <= 0) {</div> <div class="ContentPasted0">>> err = len;</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> @@ -1074,10 +1079,10 @@ int intel_migrate_copy(struct intel_migrate *m,</div> <div class="ContentPasted0">>> struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *src,</div> <div class="ContentPasted0">>> - enum i915_cache_level src_cache_level,</div> <div class="ContentPasted0">>> + unsigned int src_pat_index,</div> <div class="ContentPasted0">>> bool src_is_lmem,</div> <div class="ContentPasted0">>> struct scatterlist *dst,</div> <div class="ContentPasted0">>> - enum i915_cache_level dst_cache_level,</div> <div class="ContentPasted0">>> + unsigned int dst_pat_index,</div> <div class="ContentPasted0">>> bool dst_is_lmem,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> @@ -1098,8 +1103,8 @@ int intel_migrate_copy(struct intel_migrate *m,</div> <div class="ContentPasted0">>> goto out;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = intel_context_migrate_copy(ce, deps,</div> <div class="ContentPasted0">>> - src, src_cache_level, src_is_lmem,</div> <div class="ContentPasted0">>> - dst, dst_cache_level, dst_is_lmem,</div> <div class="ContentPasted0">>> + src, src_pat_index, src_is_lmem,</div> <div class="ContentPasted0">>> + dst, dst_pat_index, dst_is_lmem,</div> <div class="ContentPasted0">>> out);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> intel_context_unpin(ce);</div> <div class="ContentPasted0">>> @@ -1113,7 +1118,7 @@ intel_migrate_clear(struct intel_migrate *m,</div> <div class="ContentPasted0">>> struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool is_lmem,</div> <div class="ContentPasted0">>> u32 value,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> @@ -1134,7 +1139,7 @@ intel_migrate_clear(struct intel_migrate *m,</div> <div class="ContentPasted0">>> if (err)</div> <div class="ContentPasted0">>> goto out;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - err = intel_context_migrate_clear(ce, deps, sg, cache_level,</div> <div class="ContentPasted0">>> + err = intel_context_migrate_clear(ce, deps, sg, pat_index,</div> <div class="ContentPasted0">>> is_lmem, value, out);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> intel_context_unpin(ce);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.h b/drivers/gpu/drm/i915/gt/intel_migrate.h</div> <div class="ContentPasted0">>> index ccc677ec4aa3..11fc09a00c4b 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/intel_migrate.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/intel_migrate.h</div> <div class="ContentPasted0">>> @@ -16,7 +16,6 @@ struct i915_request;</div> <div class="ContentPasted0">>> struct i915_gem_ww_ctx;</div> <div class="ContentPasted0">>> struct intel_gt;</div> <div class="ContentPasted0">>> struct scatterlist;</div> <div class="ContentPasted0">>> -enum i915_cache_level;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> int intel_migrate_init(struct intel_migrate *m, struct intel_gt *gt);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -26,20 +25,20 @@ int intel_migrate_copy(struct intel_migrate *m,</div> <div class="ContentPasted0">>> struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *src,</div> <div class="ContentPasted0">>> - enum i915_cache_level src_cache_level,</div> <div class="ContentPasted0">>> + unsigned int src_pat_index,</div> <div class="ContentPasted0">>> bool src_is_lmem,</div> <div class="ContentPasted0">>> struct scatterlist *dst,</div> <div class="ContentPasted0">>> - enum i915_cache_level dst_cache_level,</div> <div class="ContentPasted0">>> + unsigned int dst_pat_index,</div> <div class="ContentPasted0">>> bool dst_is_lmem,</div> <div class="ContentPasted0">>> struct i915_request **out);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> int intel_context_migrate_copy(struct intel_context *ce,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *src,</div> <div class="ContentPasted0">>> - enum i915_cache_level src_cache_level,</div> <div class="ContentPasted0">>> + unsigned int src_pat_index,</div> <div class="ContentPasted0">>> bool src_is_lmem,</div> <div class="ContentPasted0">>> struct scatterlist *dst,</div> <div class="ContentPasted0">>> - enum i915_cache_level dst_cache_level,</div> <div class="ContentPasted0">>> + unsigned int dst_pat_index,</div> <div class="ContentPasted0">>> bool dst_is_lmem,</div> <div class="ContentPasted0">>> struct i915_request **out);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -48,7 +47,7 @@ intel_migrate_clear(struct intel_migrate *m,</div> <div class="ContentPasted0">>> struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool is_lmem,</div> <div class="ContentPasted0">>> u32 value,</div> <div class="ContentPasted0">>> struct i915_request **out);</div> <div class="ContentPasted0">>> @@ -56,7 +55,7 @@ int</div> <div class="ContentPasted0">>> intel_context_migrate_clear(struct intel_context *ce,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool is_lmem,</div> <div class="ContentPasted0">>> u32 value,</div> <div class="ContentPasted0">>> struct i915_request **out);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/intel_ppgtt.c b/drivers/gpu/drm/i915/gt/intel_ppgtt.c</div> <div class="ContentPasted0">>> index 7ecfa672f738..f0da3555c6db 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/intel_ppgtt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/intel_ppgtt.c</div> <div class="ContentPasted0">>> @@ -98,7 +98,7 @@ void</div> <div class="ContentPasted0">>> __set_pd_entry(struct i915_page_directory * const pd,</div> <div class="ContentPasted0">>> const unsigned short idx,</div> <div class="ContentPasted0">>> struct i915_page_table * const to,</div> <div class="ContentPasted0">>> - u64 (*encode)(const dma_addr_t, const enum i915_cache_level))</div> <div class="ContentPasted0">>> + u64 (*encode)(const dma_addr_t, const unsigned int))</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> /* Each thread pre-pins the pd, and we may have a thread per pde. */</div> <div class="ContentPasted0">>> GEM_BUG_ON(atomic_read(px_used(pd)) > NALLOC * I915_PDES);</div> <div class="ContentPasted0">>> @@ -181,7 +181,7 @@ struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,</div> <div class="ContentPasted0">>> void ppgtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> u32 pte_flags;</div> <div class="ContentPasted0">>> @@ -199,7 +199,7 @@ void ppgtt_bind_vma(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> if (vma_res->bi.lmem)</div> <div class="ContentPasted0">>> pte_flags |= PTE_LM;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - vm->insert_entries(vm, vma_res, cache_level, pte_flags);</div> <div class="ContentPasted0">>> + vm->insert_entries(vm, vma_res, pat_index, pte_flags);</div> <div class="ContentPasted0">>> wmb();</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/selftest_migrate.c b/drivers/gpu/drm/i915/gt/selftest_migrate.c</div> <div class="ContentPasted0">>> index e677f2da093d..3def5ca72dec 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/selftest_migrate.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/selftest_migrate.c</div> <div class="ContentPasted0">>> @@ -137,7 +137,7 @@ static int copy(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> static int intel_context_copy_ccs(struct intel_context *ce,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool write_to_ccs,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> @@ -185,7 +185,7 @@ static int intel_context_copy_ccs(struct intel_context *ce,</div> <div class="ContentPasted0">>> if (err)</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - len = emit_pte(rq, &it, cache_level, true, offset, CHUNK_SZ);</div> <div class="ContentPasted0">>> + len = emit_pte(rq, &it, pat_index, true, offset, CHUNK_SZ);</div> <div class="ContentPasted0">>> if (len <= 0) {</div> <div class="ContentPasted0">>> err = len;</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> @@ -223,7 +223,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,</div> <div class="ContentPasted0">>> struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> const struct i915_deps *deps,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool write_to_ccs,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> @@ -243,7 +243,7 @@ intel_migrate_ccs_copy(struct intel_migrate *m,</div> <div class="ContentPasted0">>> if (err)</div> <div class="ContentPasted0">>> goto out;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - err = intel_context_copy_ccs(ce, deps, sg, cache_level,</div> <div class="ContentPasted0">>> + err = intel_context_copy_ccs(ce, deps, sg, pat_index,</div> <div class="ContentPasted0">>> write_to_ccs, out);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> intel_context_unpin(ce);</div> <div class="ContentPasted0">>> @@ -300,7 +300,7 @@ static int clear(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> /* Write the obj data into ccs surface */</div> <div class="ContentPasted0">>> err = intel_migrate_ccs_copy(migrate, &ww, NULL,</div> <div class="ContentPasted0">>> obj->mm.pages->sgl,</div> <div class="ContentPasted0">>> - obj->cache_level,</div> <div class="ContentPasted0">>> + obj->pat_index,</div> <div class="ContentPasted0">>> true, &rq);</div> <div class="ContentPasted0">>> if (rq && !err) {</div> <div class="ContentPasted0">>> if (i915_request_wait(rq, 0, HZ) < 0) {</div> <div class="ContentPasted0">>> @@ -351,7 +351,7 @@ static int clear(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = intel_migrate_ccs_copy(migrate, &ww, NULL,</div> <div class="ContentPasted0">>> obj->mm.pages->sgl,</div> <div class="ContentPasted0">>> - obj->cache_level,</div> <div class="ContentPasted0">>> + obj->pat_index,</div> <div class="ContentPasted0">>> false, &rq);</div> <div class="ContentPasted0">>> if (rq && !err) {</div> <div class="ContentPasted0">>> if (i915_request_wait(rq, 0, HZ) < 0) {</div> <div class="ContentPasted0">>> @@ -414,9 +414,9 @@ static int __migrate_copy(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> return intel_migrate_copy(migrate, ww, NULL,</div> <div class="ContentPasted0">>> - src->mm.pages->sgl, src->cache_level,</div> <div class="ContentPasted0">>> + src->mm.pages->sgl, src->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(src),</div> <div class="ContentPasted0">>> - dst->mm.pages->sgl, dst->cache_level,</div> <div class="ContentPasted0">>> + dst->mm.pages->sgl, dst->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(dst),</div> <div class="ContentPasted0">>> out);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -428,9 +428,9 @@ static int __global_copy(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> struct i915_request **out)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> return intel_context_migrate_copy(migrate->context, NULL,</div> <div class="ContentPasted0">>> - src->mm.pages->sgl, src->cache_level,</div> <div class="ContentPasted0">>> + src->mm.pages->sgl, src->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(src),</div> <div class="ContentPasted0">>> - dst->mm.pages->sgl, dst->cache_level,</div> <div class="ContentPasted0">>> + dst->mm.pages->sgl, dst->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(dst),</div> <div class="ContentPasted0">>> out);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -455,7 +455,7 @@ static int __migrate_clear(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> return intel_migrate_clear(migrate, ww, NULL,</div> <div class="ContentPasted0">>> obj->mm.pages->sgl,</div> <div class="ContentPasted0">>> - obj->cache_level,</div> <div class="ContentPasted0">>> + obj->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(obj),</div> <div class="ContentPasted0">>> value, out);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -468,7 +468,7 @@ static int __global_clear(struct intel_migrate *migrate,</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> return intel_context_migrate_clear(migrate->context, NULL,</div> <div class="ContentPasted0">>> obj->mm.pages->sgl,</div> <div class="ContentPasted0">>> - obj->cache_level,</div> <div class="ContentPasted0">>> + obj->pat_index,</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(obj),</div> <div class="ContentPasted0">>> value, out);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -648,7 +648,7 @@ static int live_emit_pte_full_ring(void *arg)</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> pr_info("%s emite_pte ring space=%u\n", __func__, rq->ring->space);</div> <div class="ContentPasted0">>> it = sg_sgt(obj->mm.pages->sgl);</div> <div class="ContentPasted0">>> - len = emit_pte(rq, &it, obj->cache_level, false, 0, CHUNK_SZ);</div> <div class="ContentPasted0">>> + len = emit_pte(rq, &it, obj->pat_index, false, 0, CHUNK_SZ);</div> <div class="ContentPasted0">>> if (!len) {</div> <div class="ContentPasted0">>> err = -EINVAL;</div> <div class="ContentPasted0">>> goto out_rq;</div> <div class="ContentPasted0">>> @@ -844,7 +844,7 @@ static int wrap_ktime_compare(const void *A, const void *B)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static int __perf_clear_blt(struct intel_context *ce,</div> <div class="ContentPasted0">>> struct scatterlist *sg,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> bool is_lmem,</div> <div class="ContentPasted0">>> size_t sz)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> @@ -858,7 +858,7 @@ static int __perf_clear_blt(struct intel_context *ce,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> t0 = ktime_get();</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - err = intel_context_migrate_clear(ce, NULL, sg, cache_level,</div> <div class="ContentPasted0">>> + err = intel_context_migrate_clear(ce, NULL, sg, pat_index,</div> <div class="ContentPasted0">>> is_lmem, 0, &rq);</div> <div class="ContentPasted0">>> if (rq) {</div> <div class="ContentPasted0">>> if (i915_request_wait(rq, 0, MAX_SCHEDULE_TIMEOUT) < 0)</div> <div class="ContentPasted0">>> @@ -904,7 +904,8 @@ static int perf_clear_blt(void *arg)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = __perf_clear_blt(gt->migrate.context,</div> <div class="ContentPasted0">>> dst->mm.pages->sgl,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(dst),</div> <div class="ContentPasted0">>> sizes[i]);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -919,10 +920,10 @@ static int perf_clear_blt(void *arg)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static int __perf_copy_blt(struct intel_context *ce,</div> <div class="ContentPasted0">>> struct scatterlist *src,</div> <div class="ContentPasted0">>> - enum i915_cache_level src_cache_level,</div> <div class="ContentPasted0">>> + unsigned int src_pat_index,</div> <div class="ContentPasted0">>> bool src_is_lmem,</div> <div class="ContentPasted0">>> struct scatterlist *dst,</div> <div class="ContentPasted0">>> - enum i915_cache_level dst_cache_level,</div> <div class="ContentPasted0">>> + unsigned int dst_pat_index,</div> <div class="ContentPasted0">>> bool dst_is_lmem,</div> <div class="ContentPasted0">>> size_t sz)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> @@ -937,9 +938,9 @@ static int __perf_copy_blt(struct intel_context *ce,</div> <div class="ContentPasted0">>> t0 = ktime_get();</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = intel_context_migrate_copy(ce, NULL,</div> <div class="ContentPasted0">>> - src, src_cache_level,</div> <div class="ContentPasted0">>> + src, src_pat_index,</div> <div class="ContentPasted0">>> src_is_lmem,</div> <div class="ContentPasted0">>> - dst, dst_cache_level,</div> <div class="ContentPasted0">>> + dst, dst_pat_index,</div> <div class="ContentPasted0">>> dst_is_lmem,</div> <div class="ContentPasted0">>> &rq);</div> <div class="ContentPasted0">>> if (rq) {</div> <div class="ContentPasted0">>> @@ -994,10 +995,12 @@ static int perf_copy_blt(void *arg)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> err = __perf_copy_blt(gt->migrate.context,</div> <div class="ContentPasted0">>> src->mm.pages->sgl,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(src),</div> <div class="ContentPasted0">>> dst->mm.pages->sgl,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> i915_gem_object_is_lmem(dst),</div> <div class="ContentPasted0">>> sz);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/selftest_reset.c b/drivers/gpu/drm/i915/gt/selftest_reset.c</div> <div class="ContentPasted0">>> index a9e0a91bc0e0..79aa6ac66ad2 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/selftest_reset.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/selftest_reset.c</div> <div class="ContentPasted0">>> @@ -86,7 +86,9 @@ __igt_reset_stolen(struct intel_gt *gt,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm, dma,</div> <div class="ContentPasted0">>> ggtt->error_capture.start,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> mb();</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> s = io_mapping_map_wc(&ggtt->iomap,</div> <div class="ContentPasted0">>> @@ -127,7 +129,9 @@ __igt_reset_stolen(struct intel_gt *gt,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm, dma,</div> <div class="ContentPasted0">>> ggtt->error_capture.start,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> mb();</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> s = io_mapping_map_wc(&ggtt->iomap,</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/selftest_timeline.c b/drivers/gpu/drm/i915/gt/selftest_timeline.c</div> <div class="ContentPasted0">>> index 9f536c251179..39c3ec12df1a 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/selftest_timeline.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/selftest_timeline.c</div> <div class="ContentPasted0">>> @@ -836,7 +836,7 @@ static int setup_watcher(struct hwsp_watcher *w, struct intel_gt *gt,</div> <div class="ContentPasted0">>> return PTR_ERR(obj);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* keep the same cache settings as timeline */</div> <div class="ContentPasted0">>> - i915_gem_object_set_cache_coherency(obj, tl->hwsp_ggtt->obj->cache_level);</div> <div class="ContentPasted0">>> + i915_gem_object_set_pat_index(obj, tl->hwsp_ggtt->obj->pat_index);</div> <div class="ContentPasted0">>> w->map = i915_gem_object_pin_map_unlocked(obj,</div> <div class="ContentPasted0">>> page_unmask_bits(tl->hwsp_ggtt->obj->mm.mapping));</div> <div class="ContentPasted0">>> if (IS_ERR(w->map)) {</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/selftest_tlb.c b/drivers/gpu/drm/i915/gt/selftest_tlb.c</div> <div class="ContentPasted0">>> index e6cac1f15d6e..4493c8518e91 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/selftest_tlb.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/selftest_tlb.c</div> <div class="ContentPasted0">>> @@ -36,6 +36,8 @@ pte_tlbinv(struct intel_context *ce,</div> <div class="ContentPasted0">>> u64 length,</div> <div class="ContentPasted0">>> struct rnd_state *prng)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> + const unsigned int pat_index =</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(ce->vm->i915, I915_CACHE_NONE);</div> <div class="ContentPasted0">>> struct drm_i915_gem_object *batch;</div> <div class="ContentPasted0">>> struct drm_mm_node vb_node;</div> <div class="ContentPasted0">>> struct i915_request *rq;</div> <div class="ContentPasted0">>> @@ -155,7 +157,7 @@ pte_tlbinv(struct intel_context *ce,</div> <div class="ContentPasted0">>> /* Flip the PTE between A and B */</div> <div class="ContentPasted0">>> if (i915_gem_object_is_lmem(vb->obj))</div> <div class="ContentPasted0">>> pte_flags |= PTE_LM;</div> <div class="ContentPasted0">>> - ce->vm->insert_entries(ce->vm, &vb_res, 0, pte_flags);</div> <div class="ContentPasted0">>> + ce->vm->insert_entries(ce->vm, &vb_res, pat_index, pte_flags);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* Flush the PTE update to concurrent HW */</div> <div class="ContentPasted0">>> tlbinv(ce->vm, addr & -length, length);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c b/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c</div> <div class="ContentPasted0">>> index 6b71b9febd74..4888038c044f 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c</div> <div class="ContentPasted0">>> @@ -894,9 +894,15 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw)</div> <div class="ContentPasted0">>> pte_flags |= PTE_LM;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (ggtt->vm.raw_insert_entries)</div> <div class="ContentPasted0">>> - ggtt->vm.raw_insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags);</div> <div class="ContentPasted0">>> + ggtt->vm.raw_insert_entries(&ggtt->vm, dummy,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(ggtt->vm.i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + pte_flags);</div> <div class="ContentPasted0">>> else</div> <div class="ContentPasted0">>> - ggtt->vm.insert_entries(&ggtt->vm, dummy, I915_CACHE_NONE, pte_flags);</div> <div class="ContentPasted0">>> + ggtt->vm.insert_entries(&ggtt->vm, dummy,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(ggtt->vm.i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + pte_flags);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void uc_fw_unbind_ggtt(struct intel_uc_fw *uc_fw)</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c</div> <div class="ContentPasted0">>> index 41389a32e998..deabd69c78cc 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/i915_debugfs.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/i915_debugfs.c</div> <div class="ContentPasted0">>> @@ -139,21 +139,53 @@ static const char *stringify_vma_type(const struct i915_vma *vma)</div> <div class="ContentPasted0">>> return "ppgtt";</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> -static const char *i915_cache_level_str(struct drm_i915_private *i915, int type)</div> <div class="ContentPasted0">>> -{</div> <div class="ContentPasted0">>> - switch (type) {</div> <div class="ContentPasted0">>> - case I915_CACHE_NONE: return " uncached";</div> <div class="ContentPasted0">>> - case I915_CACHE_LLC: return HAS_LLC(i915) ? " LLC" : " snooped";</div> <div class="ContentPasted0">>> - case I915_CACHE_L3_LLC: return " L3+LLC";</div> <div class="ContentPasted0">>> - case I915_CACHE_WT: return " WT";</div> <div class="ContentPasted0">>> - default: return "";</div> <div class="ContentPasted0">>> +static const char *i915_cache_level_str(struct drm_i915_gem_object *obj)</div> <div class="ContentPasted0">>> +{</div> <div class="ContentPasted0">>> + struct drm_i915_private *i915 = obj_to_i915(obj);</div> <div class="ContentPasted0">>> +</div> <div class="ContentPasted0">>> + if (IS_METEORLAKE(i915)) {</div> <div class="ContentPasted0">>> + switch (obj->pat_index) {</div> <div class="ContentPasted0">>> + case 0: return " WB";</div> <div class="ContentPasted0">>> + case 1: return " WT";</div> <div class="ContentPasted0">>> + case 2: return " UC";</div> <div class="ContentPasted0">>> + case 3: return " WB (1-Way Coh)";</div> <div class="ContentPasted0">>> + case 4: return " WB (2-Way Coh)";</div> <div class="ContentPasted0">>> + default: return " not defined";</div> <div class="ContentPasted0">>> + }</div> <div class="ContentPasted0">>> + } else if (IS_PONTEVECCHIO(i915)) {</div> <div class="ContentPasted0">>> + switch (obj->pat_index) {</div> <div class="ContentPasted0">>> + case 0: return " UC";</div> <div class="ContentPasted0">>> + case 1: return " WC";</div> <div class="ContentPasted0">>> + case 2: return " WT";</div> <div class="ContentPasted0">>> + case 3: return " WB";</div> <div class="ContentPasted0">>> + case 4: return " WT (CLOS1)";</div> <div class="ContentPasted0">>> + case 5: return " WB (CLOS1)";</div> <div class="ContentPasted0">>> + case 6: return " WT (CLOS2)";</div> <div class="ContentPasted0">>> + case 7: return " WT (CLOS2)";</div> <div class="ContentPasted0">>> + default: return " not defined";</div> <div class="ContentPasted0">>> + }</div> <div class="ContentPasted0">>> + } else if (GRAPHICS_VER(i915) >= 12) {</div> <div class="ContentPasted0">>> + switch (obj->pat_index) {</div> <div class="ContentPasted0">>> + case 0: return " WB";</div> <div class="ContentPasted0">>> + case 1: return " WC";</div> <div class="ContentPasted0">>> + case 2: return " WT";</div> <div class="ContentPasted0">>> + case 3: return " UC";</div> <div class="ContentPasted0">>> + default: return " not defined";</div> <div class="ContentPasted0">>> + }</div> <div class="ContentPasted0">>> + } else {</div> <div class="ContentPasted0">>> + switch (obj->pat_index) {</div> <div class="ContentPasted0">>> + case 0: return " UC";</div> <div class="ContentPasted0">>> + case 1: return HAS_LLC(i915) ? " LLC" : " snooped";</div> <div class="ContentPasted0">>> + case 2: return " L3+LLC";</div> <div class="ContentPasted0">>> + case 3: return " WT";</div> <div class="ContentPasted0">>> + default: return " not defined";</div> <div class="ContentPasted0">>> + }</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> void</div> <div class="ContentPasted0">>> i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> - struct drm_i915_private *dev_priv = to_i915(obj->base.dev);</div> <div class="ContentPasted0">>> struct i915_vma *vma;</div> <div class="ContentPasted0">>> int pin_count = 0;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -165,7 +197,7 @@ i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)</div> <div class="ContentPasted0">>> obj->base.size / 1024,</div> <div class="ContentPasted0">>> obj->read_domains,</div> <div class="ContentPasted0">>> obj->write_domain,</div> <div class="ContentPasted0">>> - i915_cache_level_str(dev_priv, obj->cache_level),</div> <div class="ContentPasted0">>> + i915_cache_level_str(obj),</div> <div class="ContentPasted0">>> obj->mm.dirty ? " dirty" : "",</div> <div class="ContentPasted0">>> obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");</div> <div class="ContentPasted0">>> if (obj->base.name)</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c</div> <div class="ContentPasted0">>> index 0a78bdbd36b1..63207b0740b3 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/i915_gem.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/i915_gem.c</div> <div class="ContentPasted0">>> @@ -420,8 +420,12 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> page_length = remain < page_length ? remain : page_length;</div> <div class="ContentPasted0">>> if (drm_mm_node_allocated(&node)) {</div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm,</div> <div class="ContentPasted0">>> - i915_gem_object_get_dma_address(obj, offset >> PAGE_SHIFT),</div> <div class="ContentPasted0">>> - node.start, I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_object_get_dma_address(obj,</div> <div class="ContentPasted0">>> + offset >> PAGE_SHIFT),</div> <div class="ContentPasted0">>> + node.start,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> } else {</div> <div class="ContentPasted0">>> page_base += offset & PAGE_MASK;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> @@ -598,8 +602,12 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,</div> <div class="ContentPasted0">>> /* flush the write before we modify the GGTT */</div> <div class="ContentPasted0">>> intel_gt_flush_ggtt_writes(ggtt->vm.gt);</div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm,</div> <div class="ContentPasted0">>> - i915_gem_object_get_dma_address(obj, offset >> PAGE_SHIFT),</div> <div class="ContentPasted0">>> - node.start, I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_object_get_dma_address(obj,</div> <div class="ContentPasted0">>> + offset >> PAGE_SHIFT),</div> <div class="ContentPasted0">>> + node.start,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> wmb(); /* flush modifications to the GGTT (insert_page) */</div> <div class="ContentPasted0">>> } else {</div> <div class="ContentPasted0">>> page_base += offset & PAGE_MASK;</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c</div> <div class="ContentPasted0">>> index f020c0086fbc..2556cabea02c 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/i915_gpu_error.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c</div> <div class="ContentPasted0">>> @@ -1117,10 +1117,14 @@ i915_vma_coredump_create(const struct intel_gt *gt,</div> <div class="ContentPasted0">>> mutex_lock(&ggtt->error_mutex);</div> <div class="ContentPasted0">>> if (ggtt->vm.raw_insert_page)</div> <div class="ContentPasted0">>> ggtt->vm.raw_insert_page(&ggtt->vm, dma, slot,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> else</div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm, dma, slot,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(gt->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> mb();</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> s = io_mapping_map_wc(&ggtt->iomap, slot, PAGE_SIZE);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c</div> <div class="ContentPasted0">>> index 20a44788999e..a814775a363d 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/i915_vma.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/i915_vma.c</div> <div class="ContentPasted0">>> @@ -315,7 +315,7 @@ struct i915_vma_work {</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res;</div> <div class="ContentPasted0">>> struct drm_i915_gem_object *obj;</div> <div class="ContentPasted0">>> struct i915_sw_dma_fence_cb cb;</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level;</div> <div class="ContentPasted0">>> + unsigned int pat_index;</div> <div class="ContentPasted0">>> unsigned int flags;</div> <div class="ContentPasted0">>> };</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -334,7 +334,7 @@ static void __vma_bind(struct dma_fence_work *work)</div> <div class="ContentPasted0">>> return;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> vma_res->ops->bind_vma(vma_res->vm, &vw->stash,</div> <div class="ContentPasted0">>> - vma_res, vw->cache_level, vw->flags);</div> <div class="ContentPasted0">>> + vma_res, vw->pat_index, vw->flags);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void __vma_release(struct dma_fence_work *work)</div> <div class="ContentPasted0">>> @@ -426,7 +426,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> /**</div> <div class="ContentPasted0">>> * i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space.</div> <div class="ContentPasted0">>> * @vma: VMA to map</div> <div class="ContentPasted0">>> - * @cache_level: mapping cache level</div> <div class="ContentPasted0">>> + * @pat_index: PAT index to set in PTE</div> <div class="ContentPasted0">>> * @flags: flags like global or local mapping</div> <div class="ContentPasted0">>> * @work: preallocated worker for allocating and binding the PTE</div> <div class="ContentPasted0">>> * @vma_res: pointer to a preallocated vma resource. The resource is either</div> <div class="ContentPasted0">>> @@ -437,7 +437,7 @@ i915_vma_resource_init_from_vma(struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> * Note that DMA addresses are also the only part of the SG table we care about.</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> int i915_vma_bind(struct i915_vma *vma,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags,</div> <div class="ContentPasted0">>> struct i915_vma_work *work,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res)</div> <div class="ContentPasted0">>> @@ -507,7 +507,7 @@ int i915_vma_bind(struct i915_vma *vma,</div> <div class="ContentPasted0">>> struct dma_fence *prev;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> work->vma_res = i915_vma_resource_get(vma->resource);</div> <div class="ContentPasted0">>> - work->cache_level = cache_level;</div> <div class="ContentPasted0">>> + work->pat_index = pat_index;</div> <div class="ContentPasted0">>> work->flags = bind_flags;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /*</div> <div class="ContentPasted0">>> @@ -537,7 +537,7 @@ int i915_vma_bind(struct i915_vma *vma,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> return ret;</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> - vma->ops->bind_vma(vma->vm, NULL, vma->resource, cache_level,</div> <div class="ContentPasted0">>> + vma->ops->bind_vma(vma->vm, NULL, vma->resource, pat_index,</div> <div class="ContentPasted0">>> bind_flags);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -814,7 +814,7 @@ i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> color = 0;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (i915_vm_has_cache_coloring(vma->vm))</div> <div class="ContentPasted0">>> - color = vma->obj->cache_level;</div> <div class="ContentPasted0">>> + color = vma->obj->pat_index;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> if (flags & PIN_OFFSET_FIXED) {</div> <div class="ContentPasted0">>> u64 offset = flags & PIN_OFFSET_MASK;</div> <div class="ContentPasted0">>> @@ -1518,7 +1518,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> GEM_BUG_ON(!vma->pages);</div> <div class="ContentPasted0">>> err = i915_vma_bind(vma,</div> <div class="ContentPasted0">>> - vma->obj->cache_level,</div> <div class="ContentPasted0">>> + vma->obj->pat_index,</div> <div class="ContentPasted0">>> flags, work, vma_res);</div> <div class="ContentPasted0">>> vma_res = NULL;</div> <div class="ContentPasted0">>> if (err)</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h</div> <div class="ContentPasted0">>> index ed5c9d682a1b..31a8f8aa5558 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/i915_vma.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/i915_vma.h</div> <div class="ContentPasted0">>> @@ -250,7 +250,7 @@ i915_vma_compare(struct i915_vma *vma,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> struct i915_vma_work *i915_vma_work(void);</div> <div class="ContentPasted0">>> int i915_vma_bind(struct i915_vma *vma,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags,</div> <div class="ContentPasted0">>> struct i915_vma_work *work,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/i915_vma_types.h b/drivers/gpu/drm/i915/i915_vma_types.h</div> <div class="ContentPasted0">>> index 77fda2244d16..64472b7f0e77 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/i915_vma_types.h</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/i915_vma_types.h</div> <div class="ContentPasted0">>> @@ -32,8 +32,6 @@</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> #include "gem/i915_gem_object_types.h"</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> -enum i915_cache_level;</div> <div class="ContentPasted0">>> -</div> <div class="ContentPasted0">>> /**</div> <div class="ContentPasted0">>> * DOC: Global GTT views</div> <div class="ContentPasted0">>> *</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem.c b/drivers/gpu/drm/i915/selftests/i915_gem.c</div> <div class="ContentPasted0">>> index d91d0ade8abd..61da4ed9d521 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/selftests/i915_gem.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/selftests/i915_gem.c</div> <div class="ContentPasted0">>> @@ -57,7 +57,10 @@ static void trash_stolen(struct drm_i915_private *i915)</div> <div class="ContentPasted0">>> u32 __iomem *s;</div> <div class="ContentPasted0">>> int x;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> - ggtt->vm.insert_page(&ggtt->vm, dma, slot, I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + ggtt->vm.insert_page(&ggtt->vm, dma, slot,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> s = io_mapping_map_atomic_wc(&ggtt->iomap, slot);</div> <div class="ContentPasted0">>> for (x = 0; x < PAGE_SIZE / sizeof(u32); x++) {</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c</div> <div class="ContentPasted0">>> index f68ef4074088..f8fe3681c3dc 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c</div> <div class="ContentPasted0">>> @@ -246,7 +246,7 @@ static int igt_evict_for_cache_color(void *arg)</div> <div class="ContentPasted0">>> struct drm_mm_node target = {</div> <div class="ContentPasted0">>> .start = I915_GTT_PAGE_SIZE * 2,</div> <div class="ContentPasted0">>> .size = I915_GTT_PAGE_SIZE,</div> <div class="ContentPasted0">>> - .color = I915_CACHE_LLC,</div> <div class="ContentPasted0">>> + .color = i915_gem_get_pat_index(gt->i915, I915_CACHE_LLC),</div> <div class="ContentPasted0">>> };</div> <div class="ContentPasted0">>> struct drm_i915_gem_object *obj;</div> <div class="ContentPasted0">>> struct i915_vma *vma;</div> <div class="ContentPasted0">>> @@ -309,7 +309,7 @@ static int igt_evict_for_cache_color(void *arg)</div> <div class="ContentPasted0">>> /* Attempt to remove the first *pinned* vma, by removing the (empty)</div> <div class="ContentPasted0">>> * neighbour -- this should fail.</div> <div class="ContentPasted0">>> */</div> <div class="ContentPasted0">>> - target.color = I915_CACHE_L3_LLC;</div> <div class="ContentPasted0">>> + target.color = i915_gem_get_pat_index(gt->i915, I915_CACHE_L3_LLC);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> mutex_lock(&ggtt->vm.mutex);</div> <div class="ContentPasted0">>> err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c</div> <div class="ContentPasted0">>> index 154801f1c468..36940ef10108 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c</div> <div class="ContentPasted0">>> @@ -135,7 +135,7 @@ fake_dma_object(struct drm_i915_private *i915, u64 size)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> obj->write_domain = I915_GEM_DOMAIN_CPU;</div> <div class="ContentPasted0">>> obj->read_domains = I915_GEM_DOMAIN_CPU;</div> <div class="ContentPasted0">>> - obj->cache_level = I915_CACHE_NONE;</div> <div class="ContentPasted0">>> + obj->pat_index = i915_gem_get_pat_index(i915, I915_CACHE_NONE);</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> /* Preallocate the "backing storage" */</div> <div class="ContentPasted0">>> if (i915_gem_object_pin_pages_unlocked(obj))</div> <div class="ContentPasted0">>> @@ -359,7 +359,9 @@ static int lowlevel_hole(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> with_intel_runtime_pm(vm->gt->uncore->rpm, wakeref)</div> <div class="ContentPasted0">>> vm->insert_entries(vm, mock_vma_res,</div> <div class="ContentPasted0">>> - I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(vm->i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> count = n;</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> @@ -1377,7 +1379,10 @@ static int igt_ggtt_page(void *arg)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> ggtt->vm.insert_page(&ggtt->vm,</div> <div class="ContentPasted0">>> i915_gem_object_get_dma_address(obj, 0),</div> <div class="ContentPasted0">>> - offset, I915_CACHE_NONE, 0);</div> <div class="ContentPasted0">>> + offset,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> + 0);</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> order = i915_random_order(count, &prng);</div> <div class="ContentPasted0">>> @@ -1510,7 +1515,7 @@ static int reserve_gtt_with_resource(struct i915_vma *vma, u64 offset)</div> <div class="ContentPasted0">>> mutex_lock(&vm->mutex);</div> <div class="ContentPasted0">>> err = i915_gem_gtt_reserve(vm, NULL, &vma->node, obj->base.size,</div> <div class="ContentPasted0">>> offset,</div> <div class="ContentPasted0">>> - obj->cache_level,</div> <div class="ContentPasted0">>> + obj->pat_index,</div> <div class="ContentPasted0">>> 0);</div> <div class="ContentPasted0">>> if (!err) {</div> <div class="ContentPasted0">>> i915_vma_resource_init_from_vma(vma_res, vma);</div> <div class="ContentPasted0">>> @@ -1690,7 +1695,7 @@ static int insert_gtt_with_resource(struct i915_vma *vma)</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> mutex_lock(&vm->mutex);</div> <div class="ContentPasted0">>> err = i915_gem_gtt_insert(vm, NULL, &vma->node, obj->base.size, 0,</div> <div class="ContentPasted0">>> - obj->cache_level, 0, vm->total, 0);</div> <div class="ContentPasted0">>> + obj->pat_index, 0, vm->total, 0);</div> <div class="ContentPasted0">>> if (!err) {</div> <div class="ContentPasted0">>> i915_vma_resource_init_from_vma(vma_res, vma);</div> <div class="ContentPasted0">>> vma->resource = vma_res;</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c</div> <div class="ContentPasted0">>> index 3b18e5905c86..d985d9bae2e8 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c</div> <div class="ContentPasted0">>> @@ -1070,7 +1070,9 @@ static int igt_lmem_write_cpu(void *arg)</div> <div class="ContentPasted0">>> /* Put the pages into a known state -- from the gpu for added fun */</div> <div class="ContentPasted0">>> intel_engine_pm_get(engine);</div> <div class="ContentPasted0">>> err = intel_context_migrate_clear(engine->gt->migrate.context, NULL,</div> <div class="ContentPasted0">>> - obj->mm.pages->sgl, I915_CACHE_NONE,</div> <div class="ContentPasted0">>> + obj->mm.pages->sgl,</div> <div class="ContentPasted0">>> + i915_gem_get_pat_index(i915,</div> <div class="ContentPasted0">>> + I915_CACHE_NONE),</div> <div class="ContentPasted0">>> true, 0xdeadbeaf, &rq);</div> <div class="ContentPasted0">>> if (rq) {</div> <div class="ContentPasted0">>> dma_resv_add_fence(obj->base.resv, &rq->fence,</div> <div class="ContentPasted0">>> diff --git a/drivers/gpu/drm/i915/selftests/mock_gtt.c b/drivers/gpu/drm/i915/selftests/mock_gtt.c</div> <div class="ContentPasted0">>> index ece97e4faacb..a516c0aa88fd 100644</div> <div class="ContentPasted0">>> --- a/drivers/gpu/drm/i915/selftests/mock_gtt.c</div> <div class="ContentPasted0">>> +++ b/drivers/gpu/drm/i915/selftests/mock_gtt.c</div> <div class="ContentPasted0">>> @@ -27,21 +27,21 @@</div> <div class="ContentPasted0">>> static void mock_insert_page(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> dma_addr_t addr,</div> <div class="ContentPasted0">>> u64 offset,</div> <div class="ContentPasted0">>> - enum i915_cache_level level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void mock_insert_entries(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level level, u32 flags)</div> <div class="ContentPasted0">>> + unsigned int pat_index, u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">>> static void mock_bind_ppgtt(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> GEM_BUG_ON(flags & I915_VMA_GLOBAL_BIND);</div> <div class="ContentPasted0">>> @@ -94,7 +94,7 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name)</div> <div class="ContentPasted0">>> static void mock_bind_ggtt(struct i915_address_space *vm,</div> <div class="ContentPasted0">>> struct i915_vm_pt_stash *stash,</div> <div class="ContentPasted0">>> struct i915_vma_resource *vma_res,</div> <div class="ContentPasted0">>> - enum i915_cache_level cache_level,</div> <div class="ContentPasted0">>> + unsigned int pat_index,</div> <div class="ContentPasted0">>> u32 flags)</div> <div class="ContentPasted0">>> {</div> <div class="ContentPasted0">>> }</div> <div class="ContentPasted0">>> -- </div> <div class="ContentPasted0">>> 2.25.1</div> <div class="ContentPasted0">>> </div> <div class="ContentPasted0">></div> <div class="ContentPasted0">> -- </div> <div class="ContentPasted0">> Matt Roper</div> <div class="ContentPasted0">> Graphics Software Engineer</div> <div class="ContentPasted0">> Linux GPU Platform Enablement</div> <div class="ContentPasted0">> Intel Corporation</div> <br> </div> </body> </html>